Pileus Git - ~andy/linux/blob - net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  */
  14
  15 #include <linux/types.h>
  16 #include <linux/netfilter.h>
  17 #include <linux/module.h>
  18 #include <linux/sched.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/proc_fs.h>
  21 #include <linux/vmalloc.h>
  22 #include <linux/stddef.h>
  23 #include <linux/slab.h>
  24 #include <linux/random.h>
  25 #include <linux/jhash.h>
  26 #include <linux/err.h>
  27 #include <linux/percpu.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/notifier.h>
  30 #include <linux/kernel.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/socket.h>
  33 #include <linux/mm.h>
  34 #include <linux/nsproxy.h>
  35 #include <linux/rculist_nulls.h>
  36
  37 #include <net/netfilter/nf_conntrack.h>
  38 #include <net/netfilter/nf_conntrack_l3proto.h>
  39 #include <net/netfilter/nf_conntrack_l4proto.h>
  40 #include <net/netfilter/nf_conntrack_expect.h>
  41 #include <net/netfilter/nf_conntrack_helper.h>
  42 #include <net/netfilter/nf_conntrack_seqadj.h>
  43 #include <net/netfilter/nf_conntrack_core.h>
  44 #include <net/netfilter/nf_conntrack_extend.h>
  45 #include <net/netfilter/nf_conntrack_acct.h>
  46 #include <net/netfilter/nf_conntrack_ecache.h>
  47 #include <net/netfilter/nf_conntrack_zones.h>
  48 #include <net/netfilter/nf_conntrack_timestamp.h>
  49 #include <net/netfilter/nf_conntrack_timeout.h>
  50 #include <net/netfilter/nf_conntrack_labels.h>
  51 #include <net/netfilter/nf_conntrack_synproxy.h>
  52 #include <net/netfilter/nf_nat.h>
  53 #include <net/netfilter/nf_nat_core.h>
  54 #include <net/netfilter/nf_nat_helper.h>
  55
  56 #define NF_CONNTRACK_VERSION    "0.5.0"
  57
  58 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  59                                       enum nf_nat_manip_type manip,
  60                                       const struct nlattr *attr) __read_mostly;
  61 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  62
  63 DEFINE_SPINLOCK(nf_conntrack_lock);
  64 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  65
  66 unsigned int nf_conntrack_htable_size __read_mostly;
  67 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  68
  69 unsigned int nf_conntrack_max __read_mostly;
  70 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  71
  72 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  73 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  74
  75 unsigned int nf_conntrack_hash_rnd __read_mostly;
  76 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
  77
  78 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  79 {
  80         unsigned int n;
  81
  82         /* The direction must be ignored, so we hash everything up to the
  83          * destination ports (which is a multiple of 4) and treat the last
  84          * three bytes manually.
  85          */
  86         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  87         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  88                       (((__force __u16)tuple->dst.u.all << 16) |
  89                       tuple->dst.protonum));
  90 }
  91
  92 static u32 __hash_bucket(u32 hash, unsigned int size)
  93 {
  94         return ((u64)hash * size) >> 32;
  95 }
  96
  97 static u32 hash_bucket(u32 hash, const struct net *net)
  98 {
  99         return __hash_bucket(hash, net->ct.htable_size);
 100 }
 101
 102 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
 103                                   u16 zone, unsigned int size)
 104 {
 105         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 106 }
 107
 108 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 109                                        const struct nf_conntrack_tuple *tuple)
 110 {
 111         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 112 }
 113
 114 bool
 115 nf_ct_get_tuple(const struct sk_buff *skb,
 116                 unsigned int nhoff,
 117                 unsigned int dataoff,
 118                 u_int16_t l3num,
 119                 u_int8_t protonum,
 120                 struct nf_conntrack_tuple *tuple,
 121                 const struct nf_conntrack_l3proto *l3proto,
 122                 const struct nf_conntrack_l4proto *l4proto)
 123 {
 124         memset(tuple, 0, sizeof(*tuple));
 125
 126         tuple->src.l3num = l3num;
 127         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 128                 return false;
 129
 130         tuple->dst.protonum = protonum;
 131         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 132
 133         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 134 }
 135 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 136
 137 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 138                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 139 {
 140         struct nf_conntrack_l3proto *l3proto;
 141         struct nf_conntrack_l4proto *l4proto;
 142         unsigned int protoff;
 143         u_int8_t protonum;
 144         int ret;
 145
 146         rcu_read_lock();
 147
 148         l3proto = __nf_ct_l3proto_find(l3num);
 149         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 150         if (ret != NF_ACCEPT) {
 151                 rcu_read_unlock();
 152                 return false;
 153         }
 154
 155         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 156
 157         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 158                               l3proto, l4proto);
 159
 160         rcu_read_unlock();
 161         return ret;
 162 }
 163 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 164
 165 bool
 166 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 167                    const struct nf_conntrack_tuple *orig,
 168                    const struct nf_conntrack_l3proto *l3proto,
 169                    const struct nf_conntrack_l4proto *l4proto)
 170 {
 171         memset(inverse, 0, sizeof(*inverse));
 172
 173         inverse->src.l3num = orig->src.l3num;
 174         if (l3proto->invert_tuple(inverse, orig) == 0)
 175                 return false;
 176
 177         inverse->dst.dir = !orig->dst.dir;
 178
 179         inverse->dst.protonum = orig->dst.protonum;
 180         return l4proto->invert_tuple(inverse, orig);
 181 }
 182 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 183
 184 static void
 185 clean_from_lists(struct nf_conn *ct)
 186 {
 187         pr_debug("clean_from_lists(%p)\n", ct);
 188         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 189         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 190
 191         /* Destroy all pending expectations */
 192         nf_ct_remove_expectations(ct);
 193 }
 194
 195 static void
 196 destroy_conntrack(struct nf_conntrack *nfct)
 197 {
 198         struct nf_conn *ct = (struct nf_conn *)nfct;
 199         struct net *net = nf_ct_net(ct);
 200         struct nf_conntrack_l4proto *l4proto;
 201
 202         pr_debug("destroy_conntrack(%p)\n", ct);
 203         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 204         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 205
 206         /* To make sure we don't get any weird locking issues here:
 207          * destroy_conntrack() MUST NOT be called with a write lock
 208          * to nf_conntrack_lock!!! -HW */
 209         rcu_read_lock();
 210         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 211         if (l4proto && l4proto->destroy)
 212                 l4proto->destroy(ct);
 213
 214         rcu_read_unlock();
 215
 216         spin_lock_bh(&nf_conntrack_lock);
 217         /* Expectations will have been removed in clean_from_lists,
 218          * except TFTP can create an expectation on the first packet,
 219          * before connection is in the list, so we need to clean here,
 220          * too. */
 221         nf_ct_remove_expectations(ct);
 222
 223         /* We overload first tuple to link into unconfirmed or dying list.*/
 224         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 225         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 226
 227         NF_CT_STAT_INC(net, delete);
 228         spin_unlock_bh(&nf_conntrack_lock);
 229
 230         if (ct->master)
 231                 nf_ct_put(ct->master);
 232
 233         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 234         nf_conntrack_free(ct);
 235 }
 236
 237 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 238 {
 239         struct net *net = nf_ct_net(ct);
 240
 241         nf_ct_helper_destroy(ct);
 242         spin_lock_bh(&nf_conntrack_lock);
 243         /* Inside lock so preempt is disabled on module removal path.
 244          * Otherwise we can get spurious warnings. */
 245         NF_CT_STAT_INC(net, delete_list);
 246         clean_from_lists(ct);
 247         /* add this conntrack to the dying list */
 248         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 249                              &net->ct.dying);
 250         spin_unlock_bh(&nf_conntrack_lock);
 251 }
 252
 253 static void death_by_event(unsigned long ul_conntrack)
 254 {
 255         struct nf_conn *ct = (void *)ul_conntrack;
 256         struct net *net = nf_ct_net(ct);
 257         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 258
 259         BUG_ON(ecache == NULL);
 260
 261         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 262                 /* bad luck, let's retry again */
 263                 ecache->timeout.expires = jiffies +
 264                         (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 265                 add_timer(&ecache->timeout);
 266                 return;
 267         }
 268         /* we've got the event delivered, now it's dying */
 269         set_bit(IPS_DYING_BIT, &ct->status);
 270         nf_ct_put(ct);
 271 }
 272
 273 static void nf_ct_dying_timeout(struct nf_conn *ct)
 274 {
 275         struct net *net = nf_ct_net(ct);
 276         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 277
 278         BUG_ON(ecache == NULL);
 279
 280         /* set a new timer to retry event delivery */
 281         setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
 282         ecache->timeout.expires = jiffies +
 283                 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 284         add_timer(&ecache->timeout);
 285 }
 286
 287 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 288 {
 289         struct nf_conn_tstamp *tstamp;
 290
 291         tstamp = nf_conn_tstamp_find(ct);
 292         if (tstamp && tstamp->stop == 0)
 293                 tstamp->stop = ktime_to_ns(ktime_get_real());
 294
 295         if (!nf_ct_is_dying(ct) &&
 296             unlikely(nf_conntrack_event_report(IPCT_DESTROY, ct,
 297             portid, report) < 0)) {
 298                 /* destroy event was not delivered */
 299                 nf_ct_delete_from_lists(ct);
 300                 nf_ct_dying_timeout(ct);
 301                 return false;
 302         }
 303         set_bit(IPS_DYING_BIT, &ct->status);
 304         nf_ct_delete_from_lists(ct);
 305         nf_ct_put(ct);
 306         return true;
 307 }
 308 EXPORT_SYMBOL_GPL(nf_ct_delete);
 309
 310 static void death_by_timeout(unsigned long ul_conntrack)
 311 {
 312         nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
 313 }
 314
 315 /*
 316  * Warning :
 317  * - Caller must take a reference on returned object
 318  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 319  * OR
 320  * - Caller must lock nf_conntrack_lock before calling this function
 321  */
 322 static struct nf_conntrack_tuple_hash *
 323 ____nf_conntrack_find(struct net *net, u16 zone,
 324                       const struct nf_conntrack_tuple *tuple, u32 hash)
 325 {
 326         struct nf_conntrack_tuple_hash *h;
 327         struct hlist_nulls_node *n;
 328         unsigned int bucket = hash_bucket(hash, net);
 329
 330         /* Disable BHs the entire time since we normally need to disable them
 331          * at least once for the stats anyway.
 332          */
 333         local_bh_disable();
 334 begin:
 335         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 336                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 337                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 338                         NF_CT_STAT_INC(net, found);
 339                         local_bh_enable();
 340                         return h;
 341                 }
 342                 NF_CT_STAT_INC(net, searched);
 343         }
 344         /*
 345          * if the nulls value we got at the end of this lookup is
 346          * not the expected one, we must restart lookup.
 347          * We probably met an item that was moved to another chain.
 348          */
 349         if (get_nulls_value(n) != bucket) {
 350                 NF_CT_STAT_INC(net, search_restart);
 351                 goto begin;
 352         }
 353         local_bh_enable();
 354
 355         return NULL;
 356 }
 357
 358 /* Find a connection corresponding to a tuple. */
 359 static struct nf_conntrack_tuple_hash *
 360 __nf_conntrack_find_get(struct net *net, u16 zone,
 361                         const struct nf_conntrack_tuple *tuple, u32 hash)
 362 {
 363         struct nf_conntrack_tuple_hash *h;
 364         struct nf_conn *ct;
 365
 366         rcu_read_lock();
 367 begin:
 368         h = ____nf_conntrack_find(net, zone, tuple, hash);
 369         if (h) {
 370                 ct = nf_ct_tuplehash_to_ctrack(h);
 371                 if (unlikely(nf_ct_is_dying(ct) ||
 372                              !atomic_inc_not_zero(&ct->ct_general.use)))
 373                         h = NULL;
 374                 else {
 375                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 376                                      nf_ct_zone(ct) != zone)) {
 377                                 nf_ct_put(ct);
 378                                 goto begin;
 379                         }
 380                 }
 381         }
 382         rcu_read_unlock();
 383
 384         return h;
 385 }
 386
 387 struct nf_conntrack_tuple_hash *
 388 nf_conntrack_find_get(struct net *net, u16 zone,
 389                       const struct nf_conntrack_tuple *tuple)
 390 {
 391         return __nf_conntrack_find_get(net, zone, tuple,
 392                                        hash_conntrack_raw(tuple, zone));
 393 }
 394 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 395
 396 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 397                                        unsigned int hash,
 398                                        unsigned int repl_hash)
 399 {
 400         struct net *net = nf_ct_net(ct);
 401
 402         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 403                            &net->ct.hash[hash]);
 404         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 405                            &net->ct.hash[repl_hash]);
 406 }
 407
 408 int
 409 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 410 {
 411         struct net *net = nf_ct_net(ct);
 412         unsigned int hash, repl_hash;
 413         struct nf_conntrack_tuple_hash *h;
 414         struct hlist_nulls_node *n;
 415         u16 zone;
 416
 417         zone = nf_ct_zone(ct);
 418         hash = hash_conntrack(net, zone,
 419                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 420         repl_hash = hash_conntrack(net, zone,
 421                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 422
 423         spin_lock_bh(&nf_conntrack_lock);
 424
 425         /* See if there's one in the list already, including reverse */
 426         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 427                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 428                                       &h->tuple) &&
 429                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 430                         goto out;
 431         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 432                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 433                                       &h->tuple) &&
 434                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 435                         goto out;
 436
 437         add_timer(&ct->timeout);
 438         nf_conntrack_get(&ct->ct_general);
 439         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 440         NF_CT_STAT_INC(net, insert);
 441         spin_unlock_bh(&nf_conntrack_lock);
 442
 443         return 0;
 444
 445 out:
 446         NF_CT_STAT_INC(net, insert_failed);
 447         spin_unlock_bh(&nf_conntrack_lock);
 448         return -EEXIST;
 449 }
 450 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 451
 452 /* Confirm a connection given skb; places it in hash table */
 453 int
 454 __nf_conntrack_confirm(struct sk_buff *skb)
 455 {
 456         unsigned int hash, repl_hash;
 457         struct nf_conntrack_tuple_hash *h;
 458         struct nf_conn *ct;
 459         struct nf_conn_help *help;
 460         struct nf_conn_tstamp *tstamp;
 461         struct hlist_nulls_node *n;
 462         enum ip_conntrack_info ctinfo;
 463         struct net *net;
 464         u16 zone;
 465
 466         ct = nf_ct_get(skb, &ctinfo);
 467         net = nf_ct_net(ct);
 468
 469         /* ipt_REJECT uses nf_conntrack_attach to attach related
 470            ICMP/TCP RST packets in other direction.  Actual packet
 471            which created connection will be IP_CT_NEW or for an
 472            expected connection, IP_CT_RELATED. */
 473         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 474                 return NF_ACCEPT;
 475
 476         zone = nf_ct_zone(ct);
 477         /* reuse the hash saved before */
 478         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 479         hash = hash_bucket(hash, net);
 480         repl_hash = hash_conntrack(net, zone,
 481                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 482
 483         /* We're not in hash table, and we refuse to set up related
 484            connections for unconfirmed conns.  But packet copies and
 485            REJECT will give spurious warnings here. */
 486         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 487
 488         /* No external references means no one else could have
 489            confirmed us. */
 490         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 491         pr_debug("Confirming conntrack %p\n", ct);
 492
 493         spin_lock_bh(&nf_conntrack_lock);
 494
 495         /* We have to check the DYING flag inside the lock to prevent
 496            a race against nf_ct_get_next_corpse() possibly called from
 497            user context, else we insert an already 'dead' hash, blocking
 498            further use of that particular connection -JM */
 499
 500         if (unlikely(nf_ct_is_dying(ct))) {
 501                 spin_unlock_bh(&nf_conntrack_lock);
 502                 return NF_ACCEPT;
 503         }
 504
 505         /* See if there's one in the list already, including reverse:
 506            NAT could have grabbed it without realizing, since we're
 507            not in the hash.  If there is, we lost race. */
 508         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 509                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 510                                       &h->tuple) &&
 511                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 512                         goto out;
 513         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 514                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 515                                       &h->tuple) &&
 516                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 517                         goto out;
 518
 519         /* Remove from unconfirmed list */
 520         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 521
 522         /* Timer relative to confirmation time, not original
 523            setting time, otherwise we'd get timer wrap in
 524            weird delay cases. */
 525         ct->timeout.expires += jiffies;
 526         add_timer(&ct->timeout);
 527         atomic_inc(&ct->ct_general.use);
 528         ct->status |= IPS_CONFIRMED;
 529
 530         /* set conntrack timestamp, if enabled. */
 531         tstamp = nf_conn_tstamp_find(ct);
 532         if (tstamp) {
 533                 if (skb->tstamp.tv64 == 0)
 534                         __net_timestamp(skb);
 535
 536                 tstamp->start = ktime_to_ns(skb->tstamp);
 537         }
 538         /* Since the lookup is lockless, hash insertion must be done after
 539          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 540          * guarantee that no other CPU can find the conntrack before the above
 541          * stores are visible.
 542          */
 543         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 544         NF_CT_STAT_INC(net, insert);
 545         spin_unlock_bh(&nf_conntrack_lock);
 546
 547         help = nfct_help(ct);
 548         if (help && help->helper)
 549                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 550
 551         nf_conntrack_event_cache(master_ct(ct) ?
 552                                  IPCT_RELATED : IPCT_NEW, ct);
 553         return NF_ACCEPT;
 554
 555 out:
 556         NF_CT_STAT_INC(net, insert_failed);
 557         spin_unlock_bh(&nf_conntrack_lock);
 558         return NF_DROP;
 559 }
 560 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 561
 562 /* Returns true if a connection correspondings to the tuple (required
 563    for NAT). */
 564 int
 565 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 566                          const struct nf_conn *ignored_conntrack)
 567 {
 568         struct net *net = nf_ct_net(ignored_conntrack);
 569         struct nf_conntrack_tuple_hash *h;
 570         struct hlist_nulls_node *n;
 571         struct nf_conn *ct;
 572         u16 zone = nf_ct_zone(ignored_conntrack);
 573         unsigned int hash = hash_conntrack(net, zone, tuple);
 574
 575         /* Disable BHs the entire time since we need to disable them at
 576          * least once for the stats anyway.
 577          */
 578         rcu_read_lock_bh();
 579         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 580                 ct = nf_ct_tuplehash_to_ctrack(h);
 581                 if (ct != ignored_conntrack &&
 582                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 583                     nf_ct_zone(ct) == zone) {
 584                         NF_CT_STAT_INC(net, found);
 585                         rcu_read_unlock_bh();
 586                         return 1;
 587                 }
 588                 NF_CT_STAT_INC(net, searched);
 589         }
 590         rcu_read_unlock_bh();
 591
 592         return 0;
 593 }
 594 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 595
 596 #define NF_CT_EVICTION_RANGE    8
 597
 598 /* There's a small race here where we may free a just-assured
 599    connection.  Too bad: we're in trouble anyway. */
 600 static noinline int early_drop(struct net *net, unsigned int hash)
 601 {
 602         /* Use oldest entry, which is roughly LRU */
 603         struct nf_conntrack_tuple_hash *h;
 604         struct nf_conn *ct = NULL, *tmp;
 605         struct hlist_nulls_node *n;
 606         unsigned int i, cnt = 0;
 607         int dropped = 0;
 608
 609         rcu_read_lock();
 610         for (i = 0; i < net->ct.htable_size; i++) {
 611                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 612                                          hnnode) {
 613                         tmp = nf_ct_tuplehash_to_ctrack(h);
 614                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 615                                 ct = tmp;
 616                         cnt++;
 617                 }
 618
 619                 if (ct != NULL) {
 620                         if (likely(!nf_ct_is_dying(ct) &&
 621                                    atomic_inc_not_zero(&ct->ct_general.use)))
 622                                 break;
 623                         else
 624                                 ct = NULL;
 625                 }
 626
 627                 if (cnt >= NF_CT_EVICTION_RANGE)
 628                         break;
 629
 630                 hash = (hash + 1) % net->ct.htable_size;
 631         }
 632         rcu_read_unlock();
 633
 634         if (!ct)
 635                 return dropped;
 636
 637         if (del_timer(&ct->timeout)) {
 638                 if (nf_ct_delete(ct, 0, 0)) {
 639                         dropped = 1;
 640                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 641                 }
 642         }
 643         nf_ct_put(ct);
 644         return dropped;
 645 }
 646
 647 void init_nf_conntrack_hash_rnd(void)
 648 {
 649         unsigned int rand;
 650
 651         /*
 652          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 653          * Because there isn't enough entropy when system initializing,
 654          * and we initialize it as late as possible.
 655          */
 656         do {
 657                 get_random_bytes(&rand, sizeof(rand));
 658         } while (!rand);
 659         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 660 }
 661
 662 static struct nf_conn *
 663 __nf_conntrack_alloc(struct net *net, u16 zone,
 664                      const struct nf_conntrack_tuple *orig,
 665                      const struct nf_conntrack_tuple *repl,
 666                      gfp_t gfp, u32 hash)
 667 {
 668         struct nf_conn *ct;
 669
 670         if (unlikely(!nf_conntrack_hash_rnd)) {
 671                 init_nf_conntrack_hash_rnd();
 672                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 673                 hash = hash_conntrack_raw(orig, zone);
 674         }
 675
 676         /* We don't want any race condition at early drop stage */
 677         atomic_inc(&net->ct.count);
 678
 679         if (nf_conntrack_max &&
 680             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 681                 if (!early_drop(net, hash_bucket(hash, net))) {
 682                         atomic_dec(&net->ct.count);
 683                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
 684                         return ERR_PTR(-ENOMEM);
 685                 }
 686         }
 687
 688         /*
 689          * Do not use kmem_cache_zalloc(), as this cache uses
 690          * SLAB_DESTROY_BY_RCU.
 691          */
 692         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 693         if (ct == NULL) {
 694                 atomic_dec(&net->ct.count);
 695                 return ERR_PTR(-ENOMEM);
 696         }
 697         /*
 698          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 699          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 700          */
 701         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 702                offsetof(struct nf_conn, proto) -
 703                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 704         spin_lock_init(&ct->lock);
 705         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 706         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 707         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 708         /* save hash for reusing when confirming */
 709         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 710         /* Don't set timer yet: wait for confirmation */
 711         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 712         write_pnet(&ct->ct_net, net);
 713 #ifdef CONFIG_NF_CONNTRACK_ZONES
 714         if (zone) {
 715                 struct nf_conntrack_zone *nf_ct_zone;
 716
 717                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 718                 if (!nf_ct_zone)
 719                         goto out_free;
 720                 nf_ct_zone->id = zone;
 721         }
 722 #endif
 723         /*
 724          * changes to lookup keys must be done before setting refcnt to 1
 725          */
 726         smp_wmb();
 727         atomic_set(&ct->ct_general.use, 1);
 728         return ct;
 729
 730 #ifdef CONFIG_NF_CONNTRACK_ZONES
 731 out_free:
 732         atomic_dec(&net->ct.count);
 733         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 734         return ERR_PTR(-ENOMEM);
 735 #endif
 736 }
 737
 738 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 739                                    const struct nf_conntrack_tuple *orig,
 740                                    const struct nf_conntrack_tuple *repl,
 741                                    gfp_t gfp)
 742 {
 743         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 744 }
 745 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 746
 747 void nf_conntrack_free(struct nf_conn *ct)
 748 {
 749         struct net *net = nf_ct_net(ct);
 750
 751         nf_ct_ext_destroy(ct);
 752         nf_ct_ext_free(ct);
 753         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 754         smp_mb__before_atomic_dec();
 755         atomic_dec(&net->ct.count);
 756 }
 757 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 758
 759
 760 /* Allocate a new conntrack: we return -ENOMEM if classification
 761    failed due to stress.  Otherwise it really is unclassifiable. */
 762 static struct nf_conntrack_tuple_hash *
 763 init_conntrack(struct net *net, struct nf_conn *tmpl,
 764                const struct nf_conntrack_tuple *tuple,
 765                struct nf_conntrack_l3proto *l3proto,
 766                struct nf_conntrack_l4proto *l4proto,
 767                struct sk_buff *skb,
 768                unsigned int dataoff, u32 hash)
 769 {
 770         struct nf_conn *ct;
 771         struct nf_conn_help *help;
 772         struct nf_conntrack_tuple repl_tuple;
 773         struct nf_conntrack_ecache *ecache;
 774         struct nf_conntrack_expect *exp;
 775         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 776         struct nf_conn_timeout *timeout_ext;
 777         unsigned int *timeouts;
 778
 779         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 780                 pr_debug("Can't invert tuple.\n");
 781                 return NULL;
 782         }
 783
 784         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 785                                   hash);
 786         if (IS_ERR(ct))
 787                 return (struct nf_conntrack_tuple_hash *)ct;
 788
 789         if (tmpl && nfct_synproxy(tmpl)) {
 790                 nfct_seqadj_ext_add(ct);
 791                 nfct_synproxy_ext_add(ct);
 792         }
 793
 794         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 795         if (timeout_ext)
 796                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 797         else
 798                 timeouts = l4proto->get_timeouts(net);
 799
 800         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 801                 nf_conntrack_free(ct);
 802                 pr_debug("init conntrack: can't track with proto module\n");
 803                 return NULL;
 804         }
 805
 806         if (timeout_ext)
 807                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 808
 809         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 810         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 811         nf_ct_labels_ext_add(ct);
 812
 813         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 814         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 815                                  ecache ? ecache->expmask : 0,
 816                              GFP_ATOMIC);
 817
 818         spin_lock_bh(&nf_conntrack_lock);
 819         exp = nf_ct_find_expectation(net, zone, tuple);
 820         if (exp) {
 821                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 822                          ct, exp);
 823                 /* Welcome, Mr. Bond.  We've been expecting you... */
 824                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 825                 ct->master = exp->master;
 826                 if (exp->helper) {
 827                         help = nf_ct_helper_ext_add(ct, exp->helper,
 828                                                     GFP_ATOMIC);
 829                         if (help)
 830                                 rcu_assign_pointer(help->helper, exp->helper);
 831                 }
 832
 833 #ifdef CONFIG_NF_CONNTRACK_MARK
 834                 ct->mark = exp->master->mark;
 835 #endif
 836 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 837                 ct->secmark = exp->master->secmark;
 838 #endif
 839                 nf_conntrack_get(&ct->master->ct_general);
 840                 NF_CT_STAT_INC(net, expect_new);
 841         } else {
 842                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 843                 NF_CT_STAT_INC(net, new);
 844         }
 845
 846         /* Overload tuple linked list to put us in unconfirmed list. */
 847         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 848                        &net->ct.unconfirmed);
 849
 850         spin_unlock_bh(&nf_conntrack_lock);
 851
 852         if (exp) {
 853                 if (exp->expectfn)
 854                         exp->expectfn(ct, exp);
 855                 nf_ct_expect_put(exp);
 856         }
 857
 858         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 859 }
 860
 861 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 862 static inline struct nf_conn *
 863 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 864                   struct sk_buff *skb,
 865                   unsigned int dataoff,
 866                   u_int16_t l3num,
 867                   u_int8_t protonum,
 868                   struct nf_conntrack_l3proto *l3proto,
 869                   struct nf_conntrack_l4proto *l4proto,
 870                   int *set_reply,
 871                   enum ip_conntrack_info *ctinfo)
 872 {
 873         struct nf_conntrack_tuple tuple;
 874         struct nf_conntrack_tuple_hash *h;
 875         struct nf_conn *ct;
 876         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 877         u32 hash;
 878
 879         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 880                              dataoff, l3num, protonum, &tuple, l3proto,
 881                              l4proto)) {
 882                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 883                 return NULL;
 884         }
 885
 886         /* look for tuple match */
 887         hash = hash_conntrack_raw(&tuple, zone);
 888         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 889         if (!h) {
 890                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 891                                    skb, dataoff, hash);
 892                 if (!h)
 893                         return NULL;
 894                 if (IS_ERR(h))
 895                         return (void *)h;
 896         }
 897         ct = nf_ct_tuplehash_to_ctrack(h);
 898
 899         /* It exists; we have (non-exclusive) reference. */
 900         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 901                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 902                 /* Please set reply bit if this packet OK */
 903                 *set_reply = 1;
 904         } else {
 905                 /* Once we've had two way comms, always ESTABLISHED. */
 906                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 907                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 908                         *ctinfo = IP_CT_ESTABLISHED;
 909                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 910                         pr_debug("nf_conntrack_in: related packet for %p\n",
 911                                  ct);
 912                         *ctinfo = IP_CT_RELATED;
 913                 } else {
 914                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 915                         *ctinfo = IP_CT_NEW;
 916                 }
 917                 *set_reply = 0;
 918         }
 919         skb->nfct = &ct->ct_general;
 920         skb->nfctinfo = *ctinfo;
 921         return ct;
 922 }
 923
 924 unsigned int
 925 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 926                 struct sk_buff *skb)
 927 {
 928         struct nf_conn *ct, *tmpl = NULL;
 929         enum ip_conntrack_info ctinfo;
 930         struct nf_conntrack_l3proto *l3proto;
 931         struct nf_conntrack_l4proto *l4proto;
 932         unsigned int *timeouts;
 933         unsigned int dataoff;
 934         u_int8_t protonum;
 935         int set_reply = 0;
 936         int ret;
 937
 938         if (skb->nfct) {
 939                 /* Previously seen (loopback or untracked)?  Ignore. */
 940                 tmpl = (struct nf_conn *)skb->nfct;
 941                 if (!nf_ct_is_template(tmpl)) {
 942                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 943                         return NF_ACCEPT;
 944                 }
 945                 skb->nfct = NULL;
 946         }
 947
 948         /* rcu_read_lock()ed by nf_hook_slow */
 949         l3proto = __nf_ct_l3proto_find(pf);
 950         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 951                                    &dataoff, &protonum);
 952         if (ret <= 0) {
 953                 pr_debug("not prepared to track yet or error occurred\n");
 954                 NF_CT_STAT_INC_ATOMIC(net, error);
 955                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 956                 ret = -ret;
 957                 goto out;
 958         }
 959
 960         l4proto = __nf_ct_l4proto_find(pf, protonum);
 961
 962         /* It may be an special packet, error, unclean...
 963          * inverse of the return code tells to the netfilter
 964          * core what to do with the packet. */
 965         if (l4proto->error != NULL) {
 966                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 967                                      pf, hooknum);
 968                 if (ret <= 0) {
 969                         NF_CT_STAT_INC_ATOMIC(net, error);
 970                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 971                         ret = -ret;
 972                         goto out;
 973                 }
 974                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 975                 if (skb->nfct)
 976                         goto out;
 977         }
 978
 979         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 980                                l3proto, l4proto, &set_reply, &ctinfo);
 981         if (!ct) {
 982                 /* Not valid part of a connection */
 983                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 984                 ret = NF_ACCEPT;
 985                 goto out;
 986         }
 987
 988         if (IS_ERR(ct)) {
 989                 /* Too stressed to deal. */
 990                 NF_CT_STAT_INC_ATOMIC(net, drop);
 991                 ret = NF_DROP;
 992                 goto out;
 993         }
 994
 995         NF_CT_ASSERT(skb->nfct);
 996
 997         /* Decide what timeout policy we want to apply to this flow. */
 998         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
 999
1000         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1001         if (ret <= 0) {
1002                 /* Invalid: inverse of the return code tells
1003                  * the netfilter core what to do */
1004                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1005                 nf_conntrack_put(skb->nfct);
1006                 skb->nfct = NULL;
1007                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1008                 if (ret == -NF_DROP)
1009                         NF_CT_STAT_INC_ATOMIC(net, drop);
1010                 ret = -ret;
1011                 goto out;
1012         }
1013
1014         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1015                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1016 out:
1017         if (tmpl) {
1018                 /* Special case: we have to repeat this hook, assign the
1019                  * template again to this packet. We assume that this packet
1020                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1021                 if (ret == NF_REPEAT)
1022                         skb->nfct = (struct nf_conntrack *)tmpl;
1023                 else
1024                         nf_ct_put(tmpl);
1025         }
1026
1027         return ret;
1028 }
1029 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1030
1031 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1032                           const struct nf_conntrack_tuple *orig)
1033 {
1034         bool ret;
1035
1036         rcu_read_lock();
1037         ret = nf_ct_invert_tuple(inverse, orig,
1038                                  __nf_ct_l3proto_find(orig->src.l3num),
1039                                  __nf_ct_l4proto_find(orig->src.l3num,
1040                                                       orig->dst.protonum));
1041         rcu_read_unlock();
1042         return ret;
1043 }
1044 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1045
1046 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1047    implicitly racy: see __nf_conntrack_confirm */
1048 void nf_conntrack_alter_reply(struct nf_conn *ct,
1049                               const struct nf_conntrack_tuple *newreply)
1050 {
1051         struct nf_conn_help *help = nfct_help(ct);
1052
1053         /* Should be unconfirmed, so not in hash table yet */
1054         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1055
1056         pr_debug("Altering reply tuple of %p to ", ct);
1057         nf_ct_dump_tuple(newreply);
1058
1059         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1060         if (ct->master || (help && !hlist_empty(&help->expectations)))
1061                 return;
1062
1063         rcu_read_lock();
1064         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1065         rcu_read_unlock();
1066 }
1067 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1068
1069 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1070 void __nf_ct_refresh_acct(struct nf_conn *ct,
1071                           enum ip_conntrack_info ctinfo,
1072                           const struct sk_buff *skb,
1073                           unsigned long extra_jiffies,
1074                           int do_acct)
1075 {
1076         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1077         NF_CT_ASSERT(skb);
1078
1079         /* Only update if this is not a fixed timeout */
1080         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1081                 goto acct;
1082
1083         /* If not in hash table, timer will not be active yet */
1084         if (!nf_ct_is_confirmed(ct)) {
1085                 ct->timeout.expires = extra_jiffies;
1086         } else {
1087                 unsigned long newtime = jiffies + extra_jiffies;
1088
1089                 /* Only update the timeout if the new timeout is at least
1090                    HZ jiffies from the old timeout. Need del_timer for race
1091                    avoidance (may already be dying). */
1092                 if (newtime - ct->timeout.expires >= HZ)
1093                         mod_timer_pending(&ct->timeout, newtime);
1094         }
1095
1096 acct:
1097         if (do_acct) {
1098                 struct nf_conn_acct *acct;
1099
1100                 acct = nf_conn_acct_find(ct);
1101                 if (acct) {
1102                         struct nf_conn_counter *counter = acct->counter;
1103
1104                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1105                         atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1106                 }
1107         }
1108 }
1109 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1110
1111 bool __nf_ct_kill_acct(struct nf_conn *ct,
1112                        enum ip_conntrack_info ctinfo,
1113                        const struct sk_buff *skb,
1114                        int do_acct)
1115 {
1116         if (do_acct) {
1117                 struct nf_conn_acct *acct;
1118
1119                 acct = nf_conn_acct_find(ct);
1120                 if (acct) {
1121                         struct nf_conn_counter *counter = acct->counter;
1122
1123                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1124                         atomic64_add(skb->len - skb_network_offset(skb),
1125                                      &counter[CTINFO2DIR(ctinfo)].bytes);
1126                 }
1127         }
1128
1129         if (del_timer(&ct->timeout)) {
1130                 ct->timeout.function((unsigned long)ct);
1131                 return true;
1132         }
1133         return false;
1134 }
1135 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1136
1137 #ifdef CONFIG_NF_CONNTRACK_ZONES
1138 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1139         .len    = sizeof(struct nf_conntrack_zone),
1140         .align  = __alignof__(struct nf_conntrack_zone),
1141         .id     = NF_CT_EXT_ZONE,
1142 };
1143 #endif
1144
1145 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1146
1147 #include <linux/netfilter/nfnetlink.h>
1148 #include <linux/netfilter/nfnetlink_conntrack.h>
1149 #include <linux/mutex.h>
1150
1151 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1152  * in ip_conntrack_core, since we don't want the protocols to autoload
1153  * or depend on ctnetlink */
1154 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1155                                const struct nf_conntrack_tuple *tuple)
1156 {
1157         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1158             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1159                 goto nla_put_failure;
1160         return 0;
1161
1162 nla_put_failure:
1163         return -1;
1164 }
1165 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1166
1167 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1168         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1169         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1170 };
1171 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1172
1173 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1174                                struct nf_conntrack_tuple *t)
1175 {
1176         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1177                 return -EINVAL;
1178
1179         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1180         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1181
1182         return 0;
1183 }
1184 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1185
1186 int nf_ct_port_nlattr_tuple_size(void)
1187 {
1188         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1189 }
1190 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1191 #endif
1192
1193 /* Used by ipt_REJECT and ip6t_REJECT. */
1194 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1195 {
1196         struct nf_conn *ct;
1197         enum ip_conntrack_info ctinfo;
1198
1199         /* This ICMP is in reverse direction to the packet which caused it */
1200         ct = nf_ct_get(skb, &ctinfo);
1201         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1202                 ctinfo = IP_CT_RELATED_REPLY;
1203         else
1204                 ctinfo = IP_CT_RELATED;
1205
1206         /* Attach to new skbuff, and increment count */
1207         nskb->nfct = &ct->ct_general;
1208         nskb->nfctinfo = ctinfo;
1209         nf_conntrack_get(nskb->nfct);
1210 }
1211
1212 /* Bring out ya dead! */
1213 static struct nf_conn *
1214 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1215                 void *data, unsigned int *bucket)
1216 {
1217         struct nf_conntrack_tuple_hash *h;
1218         struct nf_conn *ct;
1219         struct hlist_nulls_node *n;
1220
1221         spin_lock_bh(&nf_conntrack_lock);
1222         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1223                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1224                         if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1225                                 continue;
1226                         ct = nf_ct_tuplehash_to_ctrack(h);
1227                         if (iter(ct, data))
1228                                 goto found;
1229                 }
1230         }
1231         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1232                 ct = nf_ct_tuplehash_to_ctrack(h);
1233                 if (iter(ct, data))
1234                         set_bit(IPS_DYING_BIT, &ct->status);
1235         }
1236         spin_unlock_bh(&nf_conntrack_lock);
1237         return NULL;
1238 found:
1239         atomic_inc(&ct->ct_general.use);
1240         spin_unlock_bh(&nf_conntrack_lock);
1241         return ct;
1242 }
1243
1244 void nf_ct_iterate_cleanup(struct net *net,
1245                            int (*iter)(struct nf_conn *i, void *data),
1246                            void *data, u32 portid, int report)
1247 {
1248         struct nf_conn *ct;
1249         unsigned int bucket = 0;
1250
1251         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1252                 /* Time to push up daises... */
1253                 if (del_timer(&ct->timeout))
1254                         nf_ct_delete(ct, portid, report);
1255
1256                 /* ... else the timer will get him soon. */
1257
1258                 nf_ct_put(ct);
1259         }
1260 }
1261 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1262
1263 static int kill_all(struct nf_conn *i, void *data)
1264 {
1265         return 1;
1266 }
1267
1268 void nf_ct_free_hashtable(void *hash, unsigned int size)
1269 {
1270         if (is_vmalloc_addr(hash))
1271                 vfree(hash);
1272         else
1273                 free_pages((unsigned long)hash,
1274                            get_order(sizeof(struct hlist_head) * size));
1275 }
1276 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1277
1278 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1279 {
1280         nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1281 }
1282 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1283
1284 static void nf_ct_release_dying_list(struct net *net)
1285 {
1286         struct nf_conntrack_tuple_hash *h;
1287         struct nf_conn *ct;
1288         struct hlist_nulls_node *n;
1289
1290         spin_lock_bh(&nf_conntrack_lock);
1291         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1292                 ct = nf_ct_tuplehash_to_ctrack(h);
1293                 /* never fails to remove them, no listeners at this point */
1294                 nf_ct_kill(ct);
1295         }
1296         spin_unlock_bh(&nf_conntrack_lock);
1297 }
1298
1299 static int untrack_refs(void)
1300 {
1301         int cnt = 0, cpu;
1302
1303         for_each_possible_cpu(cpu) {
1304                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1305
1306                 cnt += atomic_read(&ct->ct_general.use) - 1;
1307         }
1308         return cnt;
1309 }
1310
1311 void nf_conntrack_cleanup_start(void)
1312 {
1313         RCU_INIT_POINTER(ip_ct_attach, NULL);
1314 }
1315
1316 void nf_conntrack_cleanup_end(void)
1317 {
1318         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1319         while (untrack_refs() > 0)
1320                 schedule();
1321
1322 #ifdef CONFIG_NF_CONNTRACK_ZONES
1323         nf_ct_extend_unregister(&nf_ct_zone_extend);
1324 #endif
1325         nf_conntrack_proto_fini();
1326         nf_conntrack_seqadj_fini();
1327         nf_conntrack_labels_fini();
1328         nf_conntrack_helper_fini();
1329         nf_conntrack_timeout_fini();
1330         nf_conntrack_ecache_fini();
1331         nf_conntrack_tstamp_fini();
1332         nf_conntrack_acct_fini();
1333         nf_conntrack_expect_fini();
1334 }
1335
1336 /*
1337  * Mishearing the voices in his head, our hero wonders how he's
1338  * supposed to kill the mall.
1339  */
1340 void nf_conntrack_cleanup_net(struct net *net)
1341 {
1342         LIST_HEAD(single);
1343
1344         list_add(&net->exit_list, &single);
1345         nf_conntrack_cleanup_net_list(&single);
1346 }
1347
1348 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1349 {
1350         int busy;
1351         struct net *net;
1352
1353         /*
1354          * This makes sure all current packets have passed through
1355          *  netfilter framework.  Roll on, two-stage module
1356          *  delete...
1357          */
1358         synchronize_net();
1359 i_see_dead_people:
1360         busy = 0;
1361         list_for_each_entry(net, net_exit_list, exit_list) {
1362                 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1363                 nf_ct_release_dying_list(net);
1364                 if (atomic_read(&net->ct.count) != 0)
1365                         busy = 1;
1366         }
1367         if (busy) {
1368                 schedule();
1369                 goto i_see_dead_people;
1370         }
1371
1372         list_for_each_entry(net, net_exit_list, exit_list) {
1373                 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1374                 nf_conntrack_proto_pernet_fini(net);
1375                 nf_conntrack_helper_pernet_fini(net);
1376                 nf_conntrack_ecache_pernet_fini(net);
1377                 nf_conntrack_tstamp_pernet_fini(net);
1378                 nf_conntrack_acct_pernet_fini(net);
1379                 nf_conntrack_expect_pernet_fini(net);
1380                 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1381                 kfree(net->ct.slabname);
1382                 free_percpu(net->ct.stat);
1383         }
1384 }
1385
1386 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1387 {
1388         struct hlist_nulls_head *hash;
1389         unsigned int nr_slots, i;
1390         size_t sz;
1391
1392         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1393         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1394         sz = nr_slots * sizeof(struct hlist_nulls_head);
1395         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1396                                         get_order(sz));
1397         if (!hash) {
1398                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1399                 hash = vzalloc(sz);
1400         }
1401
1402         if (hash && nulls)
1403                 for (i = 0; i < nr_slots; i++)
1404                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1405
1406         return hash;
1407 }
1408 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1409
1410 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1411 {
1412         int i, bucket, rc;
1413         unsigned int hashsize, old_size;
1414         struct hlist_nulls_head *hash, *old_hash;
1415         struct nf_conntrack_tuple_hash *h;
1416         struct nf_conn *ct;
1417
1418         if (current->nsproxy->net_ns != &init_net)
1419                 return -EOPNOTSUPP;
1420
1421         /* On boot, we can set this without any fancy locking. */
1422         if (!nf_conntrack_htable_size)
1423                 return param_set_uint(val, kp);
1424
1425         rc = kstrtouint(val, 0, &hashsize);
1426         if (rc)
1427                 return rc;
1428         if (!hashsize)
1429                 return -EINVAL;
1430
1431         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1432         if (!hash)
1433                 return -ENOMEM;
1434
1435         /* Lookups in the old hash might happen in parallel, which means we
1436          * might get false negatives during connection lookup. New connections
1437          * created because of a false negative won't make it into the hash
1438          * though since that required taking the lock.
1439          */
1440         spin_lock_bh(&nf_conntrack_lock);
1441         for (i = 0; i < init_net.ct.htable_size; i++) {
1442                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1443                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1444                                         struct nf_conntrack_tuple_hash, hnnode);
1445                         ct = nf_ct_tuplehash_to_ctrack(h);
1446                         hlist_nulls_del_rcu(&h->hnnode);
1447                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1448                                                   hashsize);
1449                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1450                 }
1451         }
1452         old_size = init_net.ct.htable_size;
1453         old_hash = init_net.ct.hash;
1454
1455         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1456         init_net.ct.hash = hash;
1457         spin_unlock_bh(&nf_conntrack_lock);
1458
1459         nf_ct_free_hashtable(old_hash, old_size);
1460         return 0;
1461 }
1462 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1463
1464 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1465                   &nf_conntrack_htable_size, 0600);
1466
1467 void nf_ct_untracked_status_or(unsigned long bits)
1468 {
1469         int cpu;
1470
1471         for_each_possible_cpu(cpu)
1472                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1473 }
1474 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1475
1476 int nf_conntrack_init_start(void)
1477 {
1478         int max_factor = 8;
1479         int ret, cpu;
1480
1481         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1482          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1483         if (!nf_conntrack_htable_size) {
1484                 nf_conntrack_htable_size
1485                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1486                            / sizeof(struct hlist_head));
1487                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1488                         nf_conntrack_htable_size = 16384;
1489                 if (nf_conntrack_htable_size < 32)
1490                         nf_conntrack_htable_size = 32;
1491
1492                 /* Use a max. factor of four by default to get the same max as
1493                  * with the old struct list_heads. When a table size is given
1494                  * we use the old value of 8 to avoid reducing the max.
1495                  * entries. */
1496                 max_factor = 4;
1497         }
1498         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1499
1500         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1501                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1502                nf_conntrack_max);
1503
1504         ret = nf_conntrack_expect_init();
1505         if (ret < 0)
1506                 goto err_expect;
1507
1508         ret = nf_conntrack_acct_init();
1509         if (ret < 0)
1510                 goto err_acct;
1511
1512         ret = nf_conntrack_tstamp_init();
1513         if (ret < 0)
1514                 goto err_tstamp;
1515
1516         ret = nf_conntrack_ecache_init();
1517         if (ret < 0)
1518                 goto err_ecache;
1519
1520         ret = nf_conntrack_timeout_init();
1521         if (ret < 0)
1522                 goto err_timeout;
1523
1524         ret = nf_conntrack_helper_init();
1525         if (ret < 0)
1526                 goto err_helper;
1527
1528         ret = nf_conntrack_labels_init();
1529         if (ret < 0)
1530                 goto err_labels;
1531
1532         ret = nf_conntrack_seqadj_init();
1533         if (ret < 0)
1534                 goto err_seqadj;
1535
1536 #ifdef CONFIG_NF_CONNTRACK_ZONES
1537         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1538         if (ret < 0)
1539                 goto err_extend;
1540 #endif
1541         ret = nf_conntrack_proto_init();
1542         if (ret < 0)
1543                 goto err_proto;
1544
1545         /* Set up fake conntrack: to never be deleted, not in any hashes */
1546         for_each_possible_cpu(cpu) {
1547                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1548                 write_pnet(&ct->ct_net, &init_net);
1549                 atomic_set(&ct->ct_general.use, 1);
1550         }
1551         /*  - and look it like as a confirmed connection */
1552         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1553         return 0;
1554
1555 err_proto:
1556 #ifdef CONFIG_NF_CONNTRACK_ZONES
1557         nf_ct_extend_unregister(&nf_ct_zone_extend);
1558 err_extend:
1559 #endif
1560         nf_conntrack_seqadj_fini();
1561 err_seqadj:
1562         nf_conntrack_labels_fini();
1563 err_labels:
1564         nf_conntrack_helper_fini();
1565 err_helper:
1566         nf_conntrack_timeout_fini();
1567 err_timeout:
1568         nf_conntrack_ecache_fini();
1569 err_ecache:
1570         nf_conntrack_tstamp_fini();
1571 err_tstamp:
1572         nf_conntrack_acct_fini();
1573 err_acct:
1574         nf_conntrack_expect_fini();
1575 err_expect:
1576         return ret;
1577 }
1578
1579 void nf_conntrack_init_end(void)
1580 {
1581         /* For use by REJECT target */
1582         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1583         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1584 }
1585
1586 /*
1587  * We need to use special "null" values, not used in hash table
1588  */
1589 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1590 #define DYING_NULLS_VAL         ((1<<30)+1)
1591 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1592
1593 int nf_conntrack_init_net(struct net *net)
1594 {
1595         int ret;
1596
1597         atomic_set(&net->ct.count, 0);
1598         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1599         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1600         INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1601         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1602         if (!net->ct.stat) {
1603                 ret = -ENOMEM;
1604                 goto err_stat;
1605         }
1606
1607         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1608         if (!net->ct.slabname) {
1609                 ret = -ENOMEM;
1610                 goto err_slabname;
1611         }
1612
1613         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1614                                                         sizeof(struct nf_conn), 0,
1615                                                         SLAB_DESTROY_BY_RCU, NULL);
1616         if (!net->ct.nf_conntrack_cachep) {
1617                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1618                 ret = -ENOMEM;
1619                 goto err_cache;
1620         }
1621
1622         net->ct.htable_size = nf_conntrack_htable_size;
1623         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1624         if (!net->ct.hash) {
1625                 ret = -ENOMEM;
1626                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1627                 goto err_hash;
1628         }
1629         ret = nf_conntrack_expect_pernet_init(net);
1630         if (ret < 0)
1631                 goto err_expect;
1632         ret = nf_conntrack_acct_pernet_init(net);
1633         if (ret < 0)
1634                 goto err_acct;
1635         ret = nf_conntrack_tstamp_pernet_init(net);
1636         if (ret < 0)
1637                 goto err_tstamp;
1638         ret = nf_conntrack_ecache_pernet_init(net);
1639         if (ret < 0)
1640                 goto err_ecache;
1641         ret = nf_conntrack_helper_pernet_init(net);
1642         if (ret < 0)
1643                 goto err_helper;
1644         ret = nf_conntrack_proto_pernet_init(net);
1645         if (ret < 0)
1646                 goto err_proto;
1647         return 0;
1648
1649 err_proto:
1650         nf_conntrack_helper_pernet_fini(net);
1651 err_helper:
1652         nf_conntrack_ecache_pernet_fini(net);
1653 err_ecache:
1654         nf_conntrack_tstamp_pernet_fini(net);
1655 err_tstamp:
1656         nf_conntrack_acct_pernet_fini(net);
1657 err_acct:
1658         nf_conntrack_expect_pernet_fini(net);
1659 err_expect:
1660         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1661 err_hash:
1662         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1663 err_cache:
1664         kfree(net->ct.slabname);
1665 err_slabname:
1666         free_percpu(net->ct.stat);
1667 err_stat:
1668         return ret;
1669 }